Risk Sensitive Distributional Soft Actor Critic for Portfolio Management

Portfolio management is a critical aspect of finance, and the use of deep reinforcement learning (DRL) has shown promise in optimizing investment strategies. In this paper, we introduce a novel approach, Risk Sensitive Distributional Soft Actor-Critic (RS-DSAC) for portfolio management in the Chinese futures market. RS-DSAC leverages the power of DRL and incorporates distributional reinforcement learning and risk preferences to enhance risk-adjusted returns. We evaluate our method using data from the Chinese futures market, which presents unique challenges due to factors such as short-selling and leverage. We compare RS-DSAC with benchmark strategies, including Buy & Hold, Constant Rebalance, Soft Actor-Critic (SAC), and Distributional Soft Actor-Critic (DSAC), in terms of cumulative return, Sharpe ratio, Sortino ratio, and maximum drawdown. Our results demonstrate that RS-DSAC outperforms these strategies, achieving superior both cumulated return and risk-adjusted return. This research highlights the effectiveness of RS-DSAC in managing investment portfolio and underscores the importance of considering risk preferences in dynamic financial markets. RS-DSAC offers a promising tool for investors seeking to optimize their portfolio while effectively managing risk.